From 9af4bcb4f7698aeea589598aa31f612210ceeb4d Mon Sep 17 00:00:00 2001 From: Paul B Mahol Date: Sun, 9 Aug 2020 17:47:34 +0200 Subject: [PATCH] avcodec/cfhd: add x86 SIMD --- libavcodec/Makefile | 2 +- libavcodec/cfhd.c | 132 ++++++-------------- libavcodec/cfhd.h | 3 + libavcodec/cfhddsp.c | 57 +++++++++ libavcodec/cfhddsp.h | 36 ++++++ libavcodec/cfhdfilter.h | 67 +++++++++++ libavcodec/x86/Makefile | 2 + libavcodec/x86/cfhddsp.asm | 220 ++++++++++++++++++++++++++++++++++ libavcodec/x86/cfhddsp_init.c | 41 +++++++ 9 files changed, 461 insertions(+), 99 deletions(-) create mode 100644 libavcodec/cfhddsp.c create mode 100644 libavcodec/cfhddsp.h create mode 100644 libavcodec/cfhdfilter.h create mode 100644 libavcodec/x86/cfhddsp.asm create mode 100644 libavcodec/x86/cfhddsp_init.c diff --git a/libavcodec/Makefile b/libavcodec/Makefile index fc4294816e..26038e96d3 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -254,7 +254,7 @@ OBJS-$(CONFIG_CCAPTION_DECODER) += ccaption_dec.o ass.o OBJS-$(CONFIG_CDGRAPHICS_DECODER) += cdgraphics.o OBJS-$(CONFIG_CDTOONS_DECODER) += cdtoons.o OBJS-$(CONFIG_CDXL_DECODER) += cdxl.o -OBJS-$(CONFIG_CFHD_DECODER) += cfhd.o cfhddata.o +OBJS-$(CONFIG_CFHD_DECODER) += cfhd.o cfhddata.o cfhddsp.o OBJS-$(CONFIG_CFHD_ENCODER) += cfhdenc.o cfhddata.o OBJS-$(CONFIG_CINEPAK_DECODER) += cinepak.o OBJS-$(CONFIG_CINEPAK_ENCODER) += cinepakenc.o elbg.o diff --git a/libavcodec/cfhd.c b/libavcodec/cfhd.c index f30b50beb0..9446a18dda 100644 --- a/libavcodec/cfhd.c +++ b/libavcodec/cfhd.c @@ -36,6 +36,7 @@ #include "internal.h" #include "thread.h" #include "cfhd.h" +#include "cfhdfilter.h" #define ALPHA_COMPAND_DC_OFFSET 256 #define ALPHA_COMPAND_GAIN 9400 @@ -190,47 +191,6 @@ static inline void process_bayer(AVFrame *frame, int bpc) } } -static inline void filter(int16_t *output, ptrdiff_t out_stride, - int16_t *low, ptrdiff_t low_stride, - int16_t *high, ptrdiff_t high_stride, - int len, int clip) -{ - int16_t tmp; - int i; - - tmp = (11*low[0*low_stride] - 4*low[1*low_stride] + low[2*low_stride] + 4) >> 3; - output[(2*0+0)*out_stride] = (tmp + high[0*high_stride]) >> 1; - if (clip) - output[(2*0+0)*out_stride] = av_clip_uintp2_c(output[(2*0+0)*out_stride], clip); - - tmp = ( 5*low[0*low_stride] + 4*low[1*low_stride] - low[2*low_stride] + 4) >> 3; - output[(2*0+1)*out_stride] = (tmp - high[0*high_stride]) >> 1; - if (clip) - output[(2*0+1)*out_stride] = av_clip_uintp2_c(output[(2*0+1)*out_stride], clip); - - for (i = 1; i < len - 1; i++) { - tmp = (low[(i-1)*low_stride] - low[(i+1)*low_stride] + 4) >> 3; - output[(2*i+0)*out_stride] = (tmp + low[i*low_stride] + high[i*high_stride]) >> 1; - if (clip) - output[(2*i+0)*out_stride] = av_clip_uintp2_c(output[(2*i+0)*out_stride], clip); - - tmp = (low[(i+1)*low_stride] - low[(i-1)*low_stride] + 4) >> 3; - output[(2*i+1)*out_stride] = (tmp + low[i*low_stride] - high[i*high_stride]) >> 1; - if (clip) - output[(2*i+1)*out_stride] = av_clip_uintp2_c(output[(2*i+1)*out_stride], clip); - } - - tmp = ( 5*low[i*low_stride] + 4*low[(i-1)*low_stride] - low[(i-2)*low_stride] + 4) >> 3; - output[(2*i+0)*out_stride] = (tmp + high[i*high_stride]) >> 1; - if (clip) - output[(2*i+0)*out_stride] = av_clip_uintp2_c(output[(2*i+0)*out_stride], clip); - - tmp = (11*low[i*low_stride] - 4*low[(i-1)*low_stride] + low[(i-2)*low_stride] + 4) >> 3; - output[(2*i+1)*out_stride] = (tmp - high[i*high_stride]) >> 1; - if (clip) - output[(2*i+1)*out_stride] = av_clip_uintp2_c(output[(2*i+1)*out_stride], clip); -} - static inline void interlaced_vertical_filter(int16_t *output, int16_t *low, int16_t *high, int width, int linesize, int plane) { @@ -256,24 +216,6 @@ static inline void inverse_temporal_filter(int16_t *output, int16_t *low, int16_ } } -static void horiz_filter(int16_t *output, int16_t *low, int16_t *high, - int width) -{ - filter(output, 1, low, 1, high, 1, width, 0); -} - -static void horiz_filter_clip(int16_t *output, int16_t *low, int16_t *high, - int width, int clip) -{ - filter(output, 1, low, 1, high, 1, width, clip); -} - -static void horiz_filter_clip_bayer(int16_t *output, int16_t *low, int16_t *high, - int width, int clip) -{ - filter(output, 2, low, 1, high, 1, width, clip); -} - static void vert_filter(int16_t *output, ptrdiff_t out_stride, int16_t *low, ptrdiff_t low_stride, int16_t *high, ptrdiff_t high_stride, int len) @@ -311,6 +253,8 @@ static int alloc_buffers(AVCodecContext *avctx) return ret; avctx->pix_fmt = s->coded_format; + ff_cfhddsp_init(&s->dsp, s->bpc, avctx->pix_fmt == AV_PIX_FMT_BAYER_RGGB16); + if ((ret = av_pix_fmt_get_chroma_sub_sample(s->coded_format, &chroma_x_shift, &chroma_y_shift)) < 0) @@ -343,13 +287,13 @@ static int alloc_buffers(AVCodecContext *avctx) h2 = h4 * 2; if (s->transform_type == 0) { - s->plane[i].idwt_size = FFALIGN(height, 8) * stride; + s->plane[i].idwt_size = FFALIGN(height, 8) * stride + SUBBAND_COUNT * 64; s->plane[i].idwt_buf = av_mallocz_array(s->plane[i].idwt_size, sizeof(*s->plane[i].idwt_buf)); s->plane[i].idwt_tmp = av_malloc_array(s->plane[i].idwt_size, sizeof(*s->plane[i].idwt_tmp)); } else { - s->plane[i].idwt_size = FFALIGN(height, 8) * stride * 2; + s->plane[i].idwt_size = FFALIGN(height, 8) * stride * 2 + SUBBAND_COUNT_3D * 64; s->plane[i].idwt_buf = av_mallocz_array(s->plane[i].idwt_size, sizeof(*s->plane[i].idwt_buf)); s->plane[i].idwt_tmp = @@ -360,16 +304,16 @@ static int alloc_buffers(AVCodecContext *avctx) return AVERROR(ENOMEM); s->plane[i].subband[0] = s->plane[i].idwt_buf; - s->plane[i].subband[1] = s->plane[i].idwt_buf + 2 * w8 * h8; - s->plane[i].subband[2] = s->plane[i].idwt_buf + 1 * w8 * h8; - s->plane[i].subband[3] = s->plane[i].idwt_buf + 3 * w8 * h8; - s->plane[i].subband[4] = s->plane[i].idwt_buf + 2 * w4 * h4; - s->plane[i].subband[5] = s->plane[i].idwt_buf + 1 * w4 * h4; - s->plane[i].subband[6] = s->plane[i].idwt_buf + 3 * w4 * h4; + s->plane[i].subband[1] = s->plane[i].idwt_buf + 2 * (w8 * h8 + 64); + s->plane[i].subband[2] = s->plane[i].idwt_buf + 1 * (w8 * h8 + 64); + s->plane[i].subband[3] = s->plane[i].idwt_buf + 3 * (w8 * h8 + 64); + s->plane[i].subband[4] = s->plane[i].idwt_buf + 2 * (w4 * h4 + 64); + s->plane[i].subband[5] = s->plane[i].idwt_buf + 1 * (w4 * h4 + 64); + s->plane[i].subband[6] = s->plane[i].idwt_buf + 3 * (w4 * h4 + 64); if (s->transform_type == 0) { - s->plane[i].subband[7] = s->plane[i].idwt_buf + 2 * w2 * h2; - s->plane[i].subband[8] = s->plane[i].idwt_buf + 1 * w2 * h2; - s->plane[i].subband[9] = s->plane[i].idwt_buf + 3 * w2 * h2; + s->plane[i].subband[7] = s->plane[i].idwt_buf + 2 * (w2 * h2 + 64); + s->plane[i].subband[8] = s->plane[i].idwt_buf + 1 * (w2 * h2 + 64); + s->plane[i].subband[9] = s->plane[i].idwt_buf + 3 * (w2 * h2 + 64); } else { int16_t *frame2 = s->plane[i].subband[7] = s->plane[i].idwt_buf + 4 * w2 * h2; @@ -404,18 +348,18 @@ static int alloc_buffers(AVCodecContext *avctx) /* ll2 and ll1 commented out because they are done in-place */ s->plane[i].l_h[0] = s->plane[i].idwt_tmp; - s->plane[i].l_h[1] = s->plane[i].idwt_tmp + 2 * w8 * h8; + s->plane[i].l_h[1] = s->plane[i].idwt_tmp + 2 * (w8 * h8 + 64); // s->plane[i].l_h[2] = ll2; s->plane[i].l_h[3] = s->plane[i].idwt_tmp; - s->plane[i].l_h[4] = s->plane[i].idwt_tmp + 2 * w4 * h4; + s->plane[i].l_h[4] = s->plane[i].idwt_tmp + 2 * (w4 * h4 + 64); // s->plane[i].l_h[5] = ll1; s->plane[i].l_h[6] = s->plane[i].idwt_tmp; - s->plane[i].l_h[7] = s->plane[i].idwt_tmp + 2 * w2 * h2; + s->plane[i].l_h[7] = s->plane[i].idwt_tmp + 2 * (w2 * h2 + 64); if (s->transform_type != 0) { - int16_t *frame2 = s->plane[i].idwt_tmp + 4 * w2 * h2; + int16_t *frame2 = s->plane[i].idwt_tmp + 4 * (w2 * h2 + 64); s->plane[i].l_h[8] = frame2; - s->plane[i].l_h[9] = frame2 + 2 * w2 * h2; + s->plane[i].l_h[9] = frame2 + 2 * (w2 * h2 + 64); } } @@ -430,6 +374,7 @@ static int cfhd_decode(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *avpkt) { CFHDContext *s = avctx->priv_data; + CFHDDSPContext *dsp = &s->dsp; GetByteContext gb; ThreadFrame frame = { .f = data }; AVFrame *pic = data; @@ -1008,7 +953,7 @@ finish: high = s->plane[plane].l_h[1]; output = s->plane[plane].subband[0]; for (i = 0; i < lowpass_height * 2; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1061,7 +1006,7 @@ finish: high = s->plane[plane].l_h[4]; output = s->plane[plane].subband[0]; for (i = 0; i < lowpass_height * 2; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1128,10 +1073,7 @@ finish: } for (i = 0; i < lowpass_height * 2; i++) { - if (avctx->pix_fmt == AV_PIX_FMT_BAYER_RGGB16) - horiz_filter_clip_bayer(dst, low, high, lowpass_width, s->bpc); - else - horiz_filter_clip(dst, low, high, lowpass_width, s->bpc); + dsp->horiz_filter_clip(dst, low, high, lowpass_width, s->bpc); if (avctx->pix_fmt == AV_PIX_FMT_GBRAP12 && act_plane == 3) process_alpha(dst, lowpass_width * 2); low += lowpass_width; @@ -1145,7 +1087,7 @@ finish: high = s->plane[plane].subband[7]; output = s->plane[plane].l_h[6]; for (i = 0; i < lowpass_height; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1155,7 +1097,7 @@ finish: high = s->plane[plane].subband[9]; output = s->plane[plane].l_h[7]; for (i = 0; i < lowpass_height; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1221,7 +1163,7 @@ finish: high = s->plane[plane].l_h[1]; output = s->plane[plane].l_h[7]; for (i = 0; i < lowpass_height * 2; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1273,7 +1215,7 @@ finish: high = s->plane[plane].l_h[4]; output = s->plane[plane].l_h[7]; for (i = 0; i < lowpass_height * 2; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1310,7 +1252,7 @@ finish: high = s->plane[plane].l_h[4]; output = s->plane[plane].l_h[9]; for (i = 0; i < lowpass_height * 2; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1399,10 +1341,7 @@ finish: low = s->plane[plane].l_h[6]; high = s->plane[plane].l_h[7]; for (i = 0; i < lowpass_height * 2; i++) { - if (avctx->pix_fmt == AV_PIX_FMT_BAYER_RGGB16) - horiz_filter_clip_bayer(dst, low, high, lowpass_width, s->bpc); - else - horiz_filter_clip(dst, low, high, lowpass_width, s->bpc); + dsp->horiz_filter_clip(dst, low, high, lowpass_width, s->bpc); low += lowpass_width; high += lowpass_width; dst += dst_linesize; @@ -1413,7 +1352,7 @@ finish: high = s->plane[plane].subband[14]; output = s->plane[plane].l_h[6]; for (i = 0; i < lowpass_height; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1423,7 +1362,7 @@ finish: high = s->plane[plane].subband[16]; output = s->plane[plane].l_h[7]; for (i = 0; i < lowpass_height; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1433,7 +1372,7 @@ finish: high = s->plane[plane].subband[11]; output = s->plane[plane].l_h[8]; for (i = 0; i < lowpass_height; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1443,7 +1382,7 @@ finish: high = s->plane[plane].subband[13]; output = s->plane[plane].l_h[9]; for (i = 0; i < lowpass_height; i++) { - horiz_filter(output, low, high, lowpass_width); + dsp->horiz_filter(output, low, high, lowpass_width); low += lowpass_width; high += lowpass_width; output += lowpass_width * 2; @@ -1505,10 +1444,7 @@ finish: } for (i = 0; i < lowpass_height * 2; i++) { - if (avctx->pix_fmt == AV_PIX_FMT_BAYER_RGGB16) - horiz_filter_clip_bayer(dst, low, high, lowpass_width, s->bpc); - else - horiz_filter_clip(dst, low, high, lowpass_width, s->bpc); + dsp->horiz_filter_clip(dst, low, high, lowpass_width, s->bpc); low += lowpass_width; high += lowpass_width; dst += dst_linesize; diff --git a/libavcodec/cfhd.h b/libavcodec/cfhd.h index dc329b724b..fdc6f1e546 100644 --- a/libavcodec/cfhd.h +++ b/libavcodec/cfhd.h @@ -29,6 +29,7 @@ #include "bytestream.h" #include "get_bits.h" #include "vlc.h" +#include "cfhddsp.h" enum CFHDParam { SampleType = 1, @@ -178,6 +179,8 @@ typedef struct CFHDContext { uint8_t prescale_table[8]; Plane plane[4]; Peak peak; + + CFHDDSPContext dsp; } CFHDContext; int ff_cfhd_init_vlcs(CFHDContext *s); diff --git a/libavcodec/cfhddsp.c b/libavcodec/cfhddsp.c new file mode 100644 index 0000000000..d099eaa6d1 --- /dev/null +++ b/libavcodec/cfhddsp.c @@ -0,0 +1,57 @@ +/* + * Copyright (c) 2015-2016 Kieran Kunhya + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavutil/attributes.h" +#include "libavutil/common.h" +#include "libavutil/avassert.h" + +#include "cfhdfilter.h" +#include "cfhddsp.h" + +static void horiz_filter(int16_t *output, const int16_t *low, const int16_t *high, + int width) +{ + filter(output, 1, low, 1, high, 1, width, 0); +} + +static void horiz_filter_clip(int16_t *output, const int16_t *low, const int16_t *high, + int width, int clip) +{ + filter(output, 1, low, 1, high, 1, width, clip); +} + +static void horiz_filter_clip_bayer(int16_t *output, const int16_t *low, const int16_t *high, + int width, int clip) +{ + filter(output, 2, low, 1, high, 1, width, clip); +} + +av_cold void ff_cfhddsp_init(CFHDDSPContext *c, int depth, int bayer) +{ + c->horiz_filter = horiz_filter; + + if (bayer) + c->horiz_filter_clip = horiz_filter_clip_bayer; + else + c->horiz_filter_clip = horiz_filter_clip; + + if (ARCH_X86) + ff_cfhddsp_init_x86(c, depth, bayer); +} diff --git a/libavcodec/cfhddsp.h b/libavcodec/cfhddsp.h new file mode 100644 index 0000000000..b147972bd9 --- /dev/null +++ b/libavcodec/cfhddsp.h @@ -0,0 +1,36 @@ +/* + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_CFHDDSP_H +#define AVCODEC_CFHDDSP_H + +#include +#include + +typedef struct CFHDDSPContext { + void (*horiz_filter)(int16_t *output, const int16_t *low, const int16_t *high, + int width); + void (*horiz_filter_clip)(int16_t *output, const int16_t *low, const int16_t *high, + int width, int bpc); +} CFHDDSPContext; + +void ff_cfhddsp_init(CFHDDSPContext *c, int format, int bayer); + +void ff_cfhddsp_init_x86(CFHDDSPContext *c, int format, int bayer); + +#endif /* AVCODEC_CFHDDSP_H */ diff --git a/libavcodec/cfhdfilter.h b/libavcodec/cfhdfilter.h new file mode 100644 index 0000000000..2096660001 --- /dev/null +++ b/libavcodec/cfhdfilter.h @@ -0,0 +1,67 @@ +/* + * Copyright (c) 2015-2016 Kieran Kunhya + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_CFHDFILTER_H +#define AVCODEC_CFHDFILTER_H + +#include + +static av_always_inline void filter(int16_t *output, ptrdiff_t out_stride, + const int16_t *low, ptrdiff_t low_stride, + const int16_t *high, ptrdiff_t high_stride, + int len, int clip) +{ + int16_t tmp; + int i; + + tmp = (11*low[0*low_stride] - 4*low[1*low_stride] + low[2*low_stride] + 4) >> 3; + output[(2*0+0)*out_stride] = (tmp + high[0*high_stride]) >> 1; + if (clip) + output[(2*0+0)*out_stride] = av_clip_uintp2_c(output[(2*0+0)*out_stride], clip); + + tmp = ( 5*low[0*low_stride] + 4*low[1*low_stride] - low[2*low_stride] + 4) >> 3; + output[(2*0+1)*out_stride] = (tmp - high[0*high_stride]) >> 1; + if (clip) + output[(2*0+1)*out_stride] = av_clip_uintp2_c(output[(2*0+1)*out_stride], clip); + + for (i = 1; i < len - 1; i++) { + tmp = (low[(i-1)*low_stride] - low[(i+1)*low_stride] + 4) >> 3; + output[(2*i+0)*out_stride] = (tmp + low[i*low_stride] + high[i*high_stride]) >> 1; + if (clip) + output[(2*i+0)*out_stride] = av_clip_uintp2_c(output[(2*i+0)*out_stride], clip); + + tmp = (low[(i+1)*low_stride] - low[(i-1)*low_stride] + 4) >> 3; + output[(2*i+1)*out_stride] = (tmp + low[i*low_stride] - high[i*high_stride]) >> 1; + if (clip) + output[(2*i+1)*out_stride] = av_clip_uintp2_c(output[(2*i+1)*out_stride], clip); + } + + tmp = ( 5*low[i*low_stride] + 4*low[(i-1)*low_stride] - low[(i-2)*low_stride] + 4) >> 3; + output[(2*i+0)*out_stride] = (tmp + high[i*high_stride]) >> 1; + if (clip) + output[(2*i+0)*out_stride] = av_clip_uintp2_c(output[(2*i+0)*out_stride], clip); + + tmp = (11*low[i*low_stride] - 4*low[(i-1)*low_stride] + low[(i-2)*low_stride] + 4) >> 3; + output[(2*i+1)*out_stride] = (tmp - high[i*high_stride]) >> 1; + if (clip) + output[(2*i+1)*out_stride] = av_clip_uintp2_c(output[(2*i+1)*out_stride], clip); +} + +#endif /* AVCODEC_CFHDFILTER_H */ diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index 194135dafb..884dc0c759 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -50,6 +50,7 @@ OBJS-$(CONFIG_ADPCM_G722_ENCODER) += x86/g722dsp_init.o OBJS-$(CONFIG_ALAC_DECODER) += x86/alacdsp_init.o OBJS-$(CONFIG_APNG_DECODER) += x86/pngdsp_init.o OBJS-$(CONFIG_CAVS_DECODER) += x86/cavsdsp.o +OBJS-$(CONFIG_CFHD_DECODER) += x86/cfhddsp_init.o OBJS-$(CONFIG_DCA_DECODER) += x86/dcadsp_init.o x86/synth_filter_init.o OBJS-$(CONFIG_DNXHD_ENCODER) += x86/dnxhdenc_init.o OBJS-$(CONFIG_EXR_DECODER) += x86/exrdsp_init.o @@ -153,6 +154,7 @@ X86ASM-OBJS-$(CONFIG_ADPCM_G722_ENCODER) += x86/g722dsp.o X86ASM-OBJS-$(CONFIG_ALAC_DECODER) += x86/alacdsp.o X86ASM-OBJS-$(CONFIG_APNG_DECODER) += x86/pngdsp.o X86ASM-OBJS-$(CONFIG_CAVS_DECODER) += x86/cavsidct.o +X86ASM-OBJS-$(CONFIG_CFHD_DECODER) += x86/cfhddsp.o X86ASM-OBJS-$(CONFIG_DCA_DECODER) += x86/dcadsp.o x86/synth_filter.o X86ASM-OBJS-$(CONFIG_DIRAC_DECODER) += x86/diracdsp.o \ x86/dirac_dwt.o diff --git a/libavcodec/x86/cfhddsp.asm b/libavcodec/x86/cfhddsp.asm new file mode 100644 index 0000000000..9e1aca88d8 --- /dev/null +++ b/libavcodec/x86/cfhddsp.asm @@ -0,0 +1,220 @@ +;****************************************************************************** +;* x86-optimized functions for the CFHD decoder +;* Copyright (c) 2020 Paul B Mahol +;* +;* This file is part of FFmpeg. +;* +;* FFmpeg is free software; you can redistribute it and/or +;* modify it under the terms of the GNU Lesser General Public +;* License as published by the Free Software Foundation; either +;* version 2.1 of the License, or (at your option) any later version. +;* +;* FFmpeg is distributed in the hope that it will be useful, +;* but WITHOUT ANY WARRANTY; without even the implied warranty of +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +;* Lesser General Public License for more details. +;* +;* You should have received a copy of the GNU Lesser General Public +;* License along with FFmpeg; if not, write to the Free Software +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA +;****************************************************************************** + +%include "libavutil/x86/x86util.asm" + +SECTION_RODATA + +factor_p1_p1: dw 1, 1, 1, 1, 1, 1, 1, 1, +factor_p1_n1: dw 1, -1, 1, -1, 1, -1, 1, -1, +pd_4: times 4 dd 4 +pw_0: times 8 dw 0 +pd_m1: times 4 dd -1 +pw_1023: times 8 dw 1023 + +SECTION .text + +%macro CFHD_HORIZ_FILTER 1 +%if %1 +cglobal cfhd_horiz_filter_clip10, 5, 6, 8, output, low, high, width, bpc + DEFINE_ARGS output, low, high, width, x, temp +%else +cglobal cfhd_horiz_filter, 4, 6, 8, output, low, high, width, x + DEFINE_ARGS output, low, high, width, x, temp +%endif + shl widthd, 1 + + movsx xq, word [lowq] + imul xq, 11 + + movsx tempq, word [lowq + 2] + imul tempq, -4 + add tempq, xq + + movsx xq, word [lowq + 4] + add tempq, xq + add tempq, 4 + sar tempq, 3 + + movsx xq, word [highq] + add tempq, xq + sar tempq, 1 + +%if %1 + movd m0, tempq + CLIPW m0, [pw_0], [pw_1023] + pextrw [outputq], xm0, 0 +%else + mov word [outputq], tempw +%endif + + movsx xq, word [lowq] + imul xq, 5 + + movsx tempq, word [lowq + 2] + imul tempq, 4 + add tempq, xq + + movsx xq, word [lowq + 4] + sub tempq, xq + add tempq, 4 + sar tempq, 3 + + movsx xq, word [highq] + sub tempq, xq + sar tempq, 1 + +%if %1 + movd m0, tempq + CLIPW m0, [pw_0], [pw_1023] + pextrw [outputq + 2], xm0, 0 +%else + mov word [outputq + 2], tempw +%endif + + mov xq, 0 + +.loop: + movu m4, [lowq + xq] + movu m1, [lowq + xq + 4] + + mova m5, m4 + punpcklwd m4, m1 + punpckhwd m5, m1 + + pmaddwd m4, [factor_p1_n1] + pmaddwd m5, [factor_p1_n1] + + mova m6, m4 + mova m7, m5 + psignd m6, m6, [pd_m1] + psignd m7, m7, [pd_m1] + + paddd m4, [pd_4] + paddd m5, [pd_4] + paddd m6, [pd_4] + paddd m7, [pd_4] + + psrad m4, 3 + psrad m5, 3 + psrad m6, 3 + psrad m7, 3 + + movu m2, [lowq + xq + 2] + movu m3, [highq + xq + 2] + + mova m0, m2 + punpcklwd m2, m3 + punpckhwd m0, m3 + + mova m1, m2 + mova m3, m0 + + pmaddwd m2, [factor_p1_p1] + pmaddwd m0, [factor_p1_p1] + pmaddwd m1, [factor_p1_n1] + pmaddwd m3, [factor_p1_n1] + + paddd m2, m4 + paddd m0, m5 + paddd m1, m6 + paddd m3, m7 + + psrad m2, 1 + psrad m0, 1 + psrad m1, 1 + psrad m3, 1 + + packssdw m2, m0 + packssdw m1, m3 + + mova m0, m2 + punpcklwd m2, m1 + punpckhwd m0, m1 + +%if %1 + CLIPW m2, [pw_0], [pw_1023] + CLIPW m0, [pw_0], [pw_1023] +%endif + + movu [outputq + xq * 2 + 4], m2 + movu [outputq + xq * 2 + mmsize + 4], m0 + + add xq, mmsize + cmp xq, widthq + jl .loop + + movsx xq, word [lowq + widthq - 2] + imul xq, 5 + + movsx tempq, word [lowq + widthq - 4] + imul tempq, 4 + add tempq, xq + + movsx xq, word [lowq + widthq - 6] + sub tempq, xq + add tempq, 4 + sar tempq, 3 + + movsx xq, word [highq + widthq - 2] + add tempq, xq + sar tempq, 1 + +%if %1 + movd m0, tempq + CLIPW m0, [pw_0], [pw_1023] + pextrw [outputq + widthq * 2 - 4], xm0, 0 +%else + mov word [outputq + widthq * 2 - 4], tempw +%endif + + movsx xq, word [lowq + widthq - 2] + imul xq, 11 + + movsx tempq, word [lowq + widthq - 4] + imul tempq, -4 + add tempq, xq + + movsx xq, word [lowq + widthq - 6] + add tempq, xq + add tempq, 4 + sar tempq, 3 + + movsx xq, word [highq + widthq - 2] + sub tempq, xq + sar tempq, 1 + +%if %1 + movd m0, tempq + CLIPW m0, [pw_0], [pw_1023] + pextrw [outputq + widthq * 2 - 2], xm0, 0 +%else + mov word [outputq + widthq * 2 - 2], tempw +%endif + + RET +%endmacro + +INIT_XMM ssse3 +CFHD_HORIZ_FILTER 0 + +INIT_XMM ssse3 +CFHD_HORIZ_FILTER 1 diff --git a/libavcodec/x86/cfhddsp_init.c b/libavcodec/x86/cfhddsp_init.c new file mode 100644 index 0000000000..a7db61a344 --- /dev/null +++ b/libavcodec/x86/cfhddsp_init.c @@ -0,0 +1,41 @@ +/* + * Copyright (c) 2020 Paul B Mahol + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include + +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/x86/cpu.h" +#include "libavcodec/avcodec.h" +#include "libavcodec/cfhddsp.h" + +void ff_cfhd_horiz_filter_ssse3(int16_t *output, const int16_t *low, const int16_t *high, int width); +void ff_cfhd_horiz_filter_clip10_ssse3(int16_t *output, const int16_t *low, const int16_t *high, int width, int bpc); + +av_cold void ff_cfhddsp_init_x86(CFHDDSPContext *c, int depth, int bayer) +{ + int cpu_flags = av_get_cpu_flags(); + + if (EXTERNAL_SSSE3(cpu_flags)) { + c->horiz_filter = ff_cfhd_horiz_filter_ssse3; + if (depth == 10 && !bayer) + c->horiz_filter_clip = ff_cfhd_horiz_filter_clip10_ssse3; + } +} -- 2.26.2